New models and early failures, part 1: tested refurbished devices offer a safer, greener choice

A new device launches. A few weeks later, its maker admits a fault. Thousands of people have used a mature model for years, so buyers can see which faults appeared and which ones makers fixed. A brand-new device has no such record of use. You may pay less for the mature model, but owning it for years could still cost more than owning a new device.

The model's record cannot replace a check on the unit you buy. On refurbed, refurbished means a professional has tested the unit again, cleaned it and reconditioned it. That unit protects you from a launch fault only if someone has fixed the fault in it, or if a professional can test for the fault and the unit has passed that test. A generic 'tested' label proves nothing about a fault nobody tested for. If you buy the mature model instead of a new device and keep it for years, you save manufacturing effort and carbon.

Twelve documented launch faults show how early failures happen, what they cost in materials, carbon and money, and which checks matter before you buy.

Early failures have different causes, so they do not all point to a rushed design. A plastic strap cracks, a capacitor is fitted the wrong way round, a cooler or a power connector overheats, a batch of devices ships without some of the hardware that renders graphics, or a card reader fails. The twelve documented cases cover all of these. All twelve cases involve mass-produced devices that were already on sale and in customers' hands. They do not involve prototypes, review samples or test units.

Testing before launch cannot catch every rare fault. Take a fault that affects 1 in 10,000 units, or 0.01%. If the test always spots the fault, checking 10,000 randomly chosen units has about a 63% chance of finding at least one affected unit; checking 30,000 raises that chance to about 95%. If ten million units ship, the same fault still reaches 1,000 devices. These are calculations, not a real defect rate for any maker. Apple's 2024 longevity paper says the company tests tens of thousands of prototype devices a year. That figure is company-wide; the paper gives no sample size for any one model.

Two clocks decide how much a launch fault matters to a buyer, and they are easy to mix up.

Time since launch measures how mature the model is. The longer a model has been on sale, the more likely people are to know about its faults. Its design or production is also more likely to have changed. Each brand launches new models according to its release cycle.

Time in service measures how much use and wear an individual unit has had.

A unit of a two-year-old model can be nearly unused, while a return of a recent model can be heavily used. So a unit of a mature model is not necessarily worn out, and an old unit has not necessarily had its launch fault corrected.

A cable inside a finished product costs little. Replacing it is another matter: someone must inspect the product, open it, fit a new assembly, close it and test it again. The cable's material cost is only one part of the bill.

A fault present when a product launches adds costs in layers. First come replacement parts, followed by the work of opening, repairing, calibrating and testing the product. The logistics bill includes collection, packaging, transport and storage for the return, as well as sending the product out again. Engineers need time to find the root cause and change a supplier or a tool. Some material cannot be reused safely. Support and compensation add to the bill, as does the margin lost while sales are halted. Every new part also requires more raw-material extraction and processing.

A worked example shows the scale. These are assumed costs in US dollars, not figures taken from any company's accounts. A minor hardware repair costs $3 in parts, $17.50 for service and testing and $25 for logistics, totalling $45.50 per unit. Parts account for only about 7% of that total. If the cost ranges from $26 to $65 per unit, the bill for 10,000 units is roughly $0.26 million to $0.65 million before engineering and compensation.

Under the same assumptions, replacing a module, a complex board or a whole unit costs from about $65 to $1,010 per unit. How much of the original product remains usable depends on how easy it is to reach and repair the faulty part and on the size of the assembly being replaced.

For comparison, Counterpoint Research estimated from a teardown in 2023 that the parts in an iPhone 14 Pro Max (128 GB) cost about $464. This independent estimate excludes assembly, shipping, development and profit, and says nothing about the cost of any repair.

Public notices and company filings give a few hard numbers for three cases: two of the faults listed earlier and one additional case, the GoPro Karma. Each number has limits.

Bambu Lab A1 printers: The US Consumer Product Safety Commission's June 2024 notice covered about 12,800 printers sold in the United States for $400 to $560. Owners could choose a refund or have the heated bed and its cable replaced. At the maker's published weight of 8.3 kg per printer, the printers covered by the notice weigh about 106 tonnes in total. They needed fixing, but their combined weight does not measure waste. If every owner accepted a refund at those prices, the total would be $5.12 to $7.17 million. That is an extreme hypothetical, not a record of actual spending.

ASUS ROG Maximus Z690 Hero motherboards: The published recall notice covered about 10,000 boards listed at about $600 each. It said a reversed capacitor could cause overheating or melting. At the listed price, those boards would be worth about $6 million at retail. That figure is neither their production cost nor the cost of the recall.

GoPro Karma drone: In its 2017 annual filing, GoPro said inventory and purchase-order charges for Karma and its accessories lowered its gross margin by 180 basis points, or 1.8 percentage points. On revenue of about $1.18 billion, that gross-margin pressure amounts to roughly $21 million in lost gross profit. That figure gives a sense of the financial scale of the Karma product programme. It is not the cost of the November 2016 recall, the total refunded or a measure of how much material was destroyed.

No public dataset shows the final cost, what happened to the material and how many returns were completed for all twelve cases. Adding up the product prices or the numbers of affected products would give a false sense of precision.

Some faults create waste before a device ever reaches a customer. Manufacturers use yield to track the share of units that pass production steps. First-pass yield is the share of units that pass a step without rework, and final yield can be higher after successful rework. Neither is a failure rate after sale.

A simple calculation shows why yield matters. Suppose each production attempt costs $500, unsuccessful units are worth nothing, and the target is 100,000 good units. At a final yield of 98%, reaching the target takes about 102,041 attempts. Each good unit costs $510.20, adding roughly $1.02 million in extra costs overall. At 95%, each good unit costs $526.32, with $2.63 million in extra costs overall. At 90%, each good unit costs $555.56, with $5.56 million in extra costs overall. These are mathematical scenarios, not measured yields for any listed product.

The share of units that pass without rework falls across successive steps. If 98% of units pass each of four steps, about 92.2% get through without rework. Someone must decide whether to rework or scrap the remaining 7.8% of units; they are not automatically scrap. A strict final test also raises the recorded reject count even as it improves the products that ship. A high reject count alone says nothing about failures in customers' hands.

Accounts show cost, not kilograms. Intel's 2025 annual filing reported that its inventory reserves were $878 million higher than in 2024, mainly because of charges relating to products made during the early stages of scaling up production using its Intel 18A process. Early production can cost manufacturers more than they earn from the resulting products, but the figure does not say how much material was scrapped.

Refurbishers face the same arithmetic: 1,000 returned devices do not make 1,000 saleable ones.

It can, in four ways that offer different levels of protection, though only the first two target a particular fault.

A documented correction: The unit received the revised board, cable, housing or other remedy. This offers the strongest protection, provided the remedy works.

A fault-specific test: A check finds the defect and routes the unit to repair. Nvidia's own guidance says to check directly for missing rendering units by counting them and comparing the total with the specification.

A mature model: For a mature model, advisories and service notices are public, and you may be able to confirm whether production has been corrected. New stock of that mature model shares the same advantage. Samsung's original Galaxy Fold shows the value of waiting for a verified revision: its April 2019 failures affected review units before retail release, and the redesigned version changed the display protection and the hinge-area construction. Those early failures do not tell us how often foldables fail later in their lives.

Earlier use that screens out weak units: Some early faults show up during a device's first life, but the surviving units pick up wear and their history is unknown. NIST's reliability handbook describes early failures, a useful life and wear-out as the so-called bathtub curve. This helps explain why screening can work, but does not prove that every product follows the curve or show where one unit sits on it.

Taken together, these points show that a corrected, properly tested unit can be less likely to have one known launch fault than an uncorrected unit. That does not mean it has a lower overall risk of failure than an equivalent new device that has also been corrected. Nor does any study show that refurbished products are more reliable in every case. Manufacturers' own certified-refurbished programmes list functional tests, genuine parts where needed and a limited warranty. That describes what they promise, not a measured failure rate.

Passing a test is useful, but it tells you less than you might think. Suppose 5% of incoming units have one particular fault and the screening test never rejects a healthy unit. If the test finds none of the faulty units, 5.00% of the units that pass still have the fault. If it catches half of the faulty units, 2.56% of passing units still have the fault; if it catches 90%, that figure falls to 0.52%, and if it catches 99%, it falls to 0.053%. These are illustrative examples, not rates measured in refurbished products. They describe one fault immediately after screening, not every failure that might occur later.

So a label saying a device is fully tested matters for a particular fault only if the test can detect that fault. A display test can find dead pixels, but it cannot tell you whether the screen will resist cracking for years. A power-on test cannot confirm that a device has the specified number of rendering units, that its storage remains stable under a sustained load or that an internal safety cable is in good condition.

Even a pass on a well-designed test has its limits. A 2021 paper by Hazelwood and Pecht in IEEE Access stresses that a device's history of use may be unknown, its components may have built up stress, and pass-or-fail checks during refurbishment have limits. Replacing a battery does not reset the age of every solder joint, connector, display or chip. Repair also raises questions. Were the right parts used and the work done properly? Was the device calibrated, sealed and given a final check?

A cosmetic condition grade does not answer these questions. It describes appearance, not whether a recall remedy was completed or how much wear there is inside. The condition grades explain what each grade covers.

For a safety-related fault, documented completion of the remedy is essential, and ordinary functional testing cannot replace it. A returned product whose recall is still unresolved should go through the correction process, not be put up for sale.

Four studies provide most of the evidence here, but each has limits.

ADEME: The French environment agency compared refurbished products with new equivalents, measuring their environmental impacts per year of use. It found that these impacts were 64 to 87% lower for refurbished smartphones, 46 to 80% lower for tablets and 43 to 97% lower for both laptops and desktop computers. The ranges cover several kinds of environmental impact under different modelled conditions, but exclude ionising radiation. They do not give a carbon-footprint reduction that applies everywhere. ADEME's 2022 briefing also estimates savings for each year a refurbished device is used. For a smartphone, it estimates about 76.9 kg less raw-material extraction and a reduction in climate impact of about 24.6 kg CO₂e. For a tablet, it estimates about 80 kg less raw-material extraction and a reduction in climate impact of about 20 kg CO₂e. For a laptop, it estimates about 127 kg less raw-material extraction and a reduction in climate impact of about 27 kg CO₂e. These are modelled savings from stages before a device is used, not measurements of its weight.

Fraunhofer Austria: A 2023 study commissioned by refurbed modelled the iPhone 11's carbon footprint at about 72 kg CO₂ when new and about 15.7 kg when refurbished, roughly 78% lower. The result comes from a model, not an audit of any listing or purchase.

The Joint Research Centre: A life-cycle study led by the European Commission's Joint Research Centre found that keeping a smartphone for three years instead of two lowers its carbon footprint by 23 to 30% in its modelled European scenarios. The benefit was greatest when each reused device replaced one new device.

A January 2026 study in Communications Earth & Environment: Using US survey data, the study estimated that each second-hand smartphone sale displaced an average of 0.40 new-device sales. It also estimated that buying and using smartphones in ways that keep them in circulation had a 34% lower annual carbon footprint. Shorter use times cancelled nearly half of the potential gain. These are estimates for second-hand smartphones in the US, not figures that apply everywhere.

A 2026 review of fifteen studies in the Journal of Remanufacturing found substantial environmental benefits, but the studies did not all measure the same stages of a device's life. Seven left out the impact of using the device. refurbed explains its own approach on its sustainability page.

Buying a refurbished device avoids the impact of making a new one only if it replaces a new purchase. Some buyers would not have bought a new device anyway, so a refurbished purchase replaces less than one new-device purchase on average. Consider an example with round numbers: making and delivering a new device causes 100 kg CO₂e of emissions, refurbishing a device causes 10 kg CO₂e, and both devices have the same impact during use. If the refurbished device replaces one new-device purchase, it avoids 90 kg CO₂e. If a refurbished purchase replaces half a new-device purchase on average, it avoids 40 kg CO₂e. If it replaces 0.4 of a new-device purchase, it avoids 30 kg CO₂e; if it replaces 0.2, it avoids 10 kg CO₂e; and if it replaces 0.1, it avoids nothing. If it replaces no new purchase, it adds 10 kg CO₂e.

The 0.4 figure reflects a 2026 US estimate for second-hand phones. Here, it is one possible replacement rate in the example, not a rule for every product.

How long a device lasts can make as much difference to the calculation as whether it replaces a new purchase. If a new device has a total environmental impact of 100 units across four years of use, that works out at 25 units a year. A refurbished device with a total impact of 20 units works out at about 6.7 units a year if used for three years, 20 units a year if used for one year and 40 units a year if used for six months. The refurbished device needs to last more than about 0.8 years to have a lower annual impact than the new device's 25 units a year. This comparison does not account for differences in the environmental impact of using the two devices.

ADEME's briefing warns that if refurbishing a laptop requires extensive replacement of its screen, memory and storage with new parts but extends its life only briefly, the laptop can have a higher environmental impact than a new device used for five years.

For energy-hungry devices, you also need to count the impact of the electricity they use. A computer, graphics card or printer that uses 40 W more over 3,000 hours consumes an extra 120 kWh. At 0.05 kg CO₂e per kWh, that adds 6 kg CO₂e. At 0.40 kg CO₂e per kWh, it adds 48 kg CO₂e; at 0.80 kg CO₂e per kWh, it adds 96 kg CO₂e. Carbon emissions from electricity vary by country, so these examples do not assume any particular location.

Two more situations can erase the benefit. Buying a refurbished device as a spare because it is affordable adds environmental impact but does little to avoid making a new one. A device left in a drawer never gets used. If the device you own already does the job, keeping it is the lowest-impact option. The environmental case is strongest when a refurbished device replaces a new purchase and stays in use for years.

Compare a refurbished device with a current new device that does the same job, using today's prices rather than the new device's launch price. To work out the cost of owning a device, start with its purchase price. Add any spending on a battery, parts, repairs and energy, as well as the cost of downtime and the effort of sorting out problems. Then subtract what you get when you sell it.

Here is an example using round numbers over four years, with a comparable new device priced at 100.

New device: You pay 100 to buy it and 3 for repairs, while downtime costs 2. You get 20 when you sell it afterwards, bringing the cost of ownership to 85.

Strong refurbished offer: You pay 60 to buy it, 5 for a battery or parts and 5 for repairs, while downtime costs 3. You get 10 when you sell it afterwards, bringing the cost of ownership to 63, about 26% less than for the new device.

Central case: You pay 70 to buy it, 8 for a battery or parts and 5 for repairs, while downtime costs 3. You get 10 when you sell it afterwards, bringing the cost of ownership to 76, about 11% less than for the new device.

Weak offer: You pay 85 to buy it, 8 for a battery or parts and 7 for repairs, while downtime costs 4. You get 10 when you sell it afterwards, bringing the cost of ownership to 94, about 11% more than for the new device.

So a purchase price that is 30% lower can mean a saving of only about 11% overall, and a smaller discount can vanish. These are illustrations, not observed discounts, repair rates or resale values.

A break-even purchase price is a better guide than a fixed discount. In the central case, you allow 8 for a battery or parts, 5 for repairs and 3 for downtime. Subtract the 10 you get when you sell it, and the costs beyond the purchase price come to 6 units. To break even with the new device, you can pay up to 85 minus 6, or 79, for the refurbished one. At 70, you save 9 units. If energy, repairs or downtime add another 10 units, that break-even price falls to 69. If the refurbished device lasts three years instead of four, its ownership cost of 76 works out at about 25.3 a year, compared with about 21.3 a year for the new device's ownership cost of 85 over four years.

Real repair prices show how fast a discount can disappear. A 2023 European Commission impact assessment sampled smartphone repair prices charged by the original manufacturers. It found that a battery replacement cost about 14% of the average new-phone price and a display replacement about 42%. With a 30% saving on the purchase price, a battery replacement you pay for yourself reduces your saving to about 16% of the new device's price. If you pay for a display replacement instead, you spend about 12% more than the new device's price. These figures leave out resale value, cover and running costs.

A warranty makes the seller responsible for certain costs if something goes wrong; it does not make hardware fail less often. A replaceable battery, available parts and a long software-support window help determine how long the device stays useful.

Are refurbished devices more reliable than new ones?

Not as a rule. A unit that has had its launch fault fixed and has been tested for that fault may be less likely to suffer from it. But no study shows that refurbished devices are more reliable overall, and previous use adds wear. Look at the tests carried out on each unit.

Is it better to wait before buying a new release?

Often, yes. In the twelve documented cases, the fault became public between about two weeks and four months after launch, and fixes followed. A mature model with the fix already applied is the safer buy; the release cycle shows when each brand's next model arrives.

Does buying refurbished save carbon?

Yes, when it replaces a new purchase and you keep it for years. The French agency ADEME found that, per year of use, the environmental impacts of refurbished smartphones were 64 to 87% lower than those of new ones. A spare device or a short second life shrinks the saving.

When a refurbished device is labelled as tested, what has been checked?

Only what the test checks. Ask which functions were tested, whether the device was checked for the model's known fault or had the fix applied, and what condition the battery is in. Switching a device on cannot reveal a fault that has not yet shown any signs, so compare the quality checks behind each offer.